Bubble Boy

If it feels like one, that’s because it is

That’s the thing about bubbles, they creep up on you. Bubbles turn usually sensible, risk-averse people into brazen speculators. This is because bubbles are always based on a strand of truth, a thread of sensibility. It’s only in hindsight, after the bubble has popped, that the absurdity becomes obvious.

Every era has a company or thing that best embodies the bubble. During the tulip bubble it was the Semper Augustus bulb (which at one point sold for the price of a grand Amsterdam mansion); the Roaring 20s had RCA, the entertainment business that later created NBC and whose share price would subsequently drop 98%.

The 1960s had the Nifty Fifty and Polaroid which had a near 100 price earnings multiple (before crashing 90%). The Japanese Stock Bubble of the 1980s, which saw its market peak in 1989 before collapsing, took until 2024 to regain its losses (meanwhile the Japanese property market never fully recovered). The star back then was NTT Docomo, which traded on a PE multiple of more than 350 and was worth more than IBM, AT&T, Exxon, General Electric, and General Motors combined (Docomo was eventually delisted in 2020 after being absorbed by its parent, NTT).

While many point to Webvan and Pets.com during the dot-com era, they were largely irrelevant – Cisco was the epoch. It was briefly the world’s most valuable company after its share price surged by over 3,800% in five years. Shortly before the dot-com bubble burst in late March 2000, Cisco snatched the crown from Microsoft, worth more than US$500 billion with a PE multiple of 230. Its share price would soon drop by 80%. It took 25 years to recover those nominal share price losses.

All these bubble epoch companies had genuine reasons for their rise; virtually all survived the crash for decades to come, albeit with far lower market values. RCA was purchased in 1985 by a then-dominant General Electric (itself a key figure in the dot-com bubble based on the now widely discredited Jack Welch’s fraudulent use of GE Capital).

What is the AI era’s poster child for the bubble?

It’s not Nvidia or Apple or Alphabet (hugely profitable businesses albeit all mutually inflated by off balance sheet capex).

No, it’s SpaceX (and when the inevitable Tesla merger happens, Elon Inc).

SpaceX has it all – a genius founder and CEO (now the world’s first and only trillionaire), a plausible bubble story (mining space) and a transformational product (Starlink).

Real bubbles need smart people to join in, not just brazen speculators. The NFT craze wasn’t a real bubble, only a handful of crazed speculators, pop stars and late-to-the-party idiots got involved. The smart money stayed well clear.

SpaceX is different. Australia’s smartest debt fund manager, Chris Joye, claimed SpaceX’s “equity story was all about the almost unquantifiable optionality and upside associated with the world’s most capable, and important, manufacturing business.” Marc Andreeson, founder of Netscape and one of the most influential VCs in the Valley wrote a fawning profile outlining the upside of SpaceX’s moon shots (while assiduously avoiding any real financial metrics).

Australia’s richest person, Gina Rinehart, invested $1.4 billion in the IPO. More than 20 of Wall Street’s most respected banks worked on the float, which was led by Goldman Sachs and Morgan Stanley.

Even Joye fell into the classic bubble trap, misplacing product genius with positive cash flows: “SpaceX inverted the economics. It used vertical integration, a never-before-seen launch cadence, and reusable rocketry to turn a failed industry into a planetary utility. Starlink now connects households, airlines, ships, emergency services and armies. In Ukraine, access to Starlink materially changed the battlefield.”

To be clear, I’m no Elon hater – I drive a Tesla (my favourite ever car) and love Starlink. We once recorded a podcast from midair using Starlink with the full episode uploaded by the time I landed. The SpaceX rocket business has a clear lead over all-comers which will persist for decades.

The problem? Having a great product and visionary genius doesn’t justify a price to sales multiple of 100 (or a forward P/S multiple of 90). Bearing in mind we shouldn’t even be using the insidious revenue multiple, we just have no choice here because SpaceX not only doesn’t make any money, it won’t make any real money (post capex) for a long time.

The intrinsic value of SpaceX is so far removed from its current valuation (around $2.1 trillion depending on the day) that it’s difficult to take any of this too seriously. But let’s do a quick sum of the parts given that SpaceX is really two (or possibly three) separate businesses.

First the easy one – xAI. Elon’s bastardised social media business turned frontier model turned data centre operator.

xAI is a cash furnace. Last year it saw US$1.2b in EBITDA losses before its capex of US$12b. Even with the recent deals to rent GPUs to Anthropic and Google, this is a business that is worth essentially nothing and competing against Anthropic and OpenAI (and a host of other models). The Cursor deal (which got significantly cheaper than the US$60b sticker price with SpaceX’s share price gain) is interesting, but the LLM space is becoming commoditised and margins are likely to be starched away through furious competition between frontier labs and lightweight models. (Note, the cash burn of xAI will drop materially following the Anthropic and Google deals but then it becomes more of a GPU rental business than a deep tech decacorn).

Then there’s the good business – Starlink. Starlink has a virtual monopoly on satellite internet globally (Amazon’s Leo is miles behind). It benefits from being able to hitch a ride on SpaceX’s launch capability for essentially a wholesale price.

There are around 10,000 Starlink satellites in orbit out of 15,400 total satellites – Starlink is literally two thirds of all low orbit satellites. That’s a big lead that could take a decade or more to catch. When Starship ramps up the cost per kilogram will drop 90% and that benefit will continue to be mostly capured by Starlink.

But even virtual monpolies have an intrinsic value. For that we need to look at how much real profit (after capex) the business generates, how fast its profit is growing and how big is its addressable market.

According to SpaceX’s S1 Starlink makes around US$8 billion a year in EBITDA. But this isn’t a media or software business with no incremental capital costs – satellites aren’t cheap, even the little ones. The business spent US$4b on capital expenditures – so a better proxy for profit is US$4 billion.

The strange thing is, despite Starlink being such an incredible product, it’s not actually growing that fast, it rose around 86 percent last year. This is very good, but it’s not 100x sales good.

Meanwhile, the average revenue per user for Starlink is dropping – from US$99 per month in 2023 to US$81 per month in 2025. This reflects the shift from high cost users like private jets and super yachts to cheaper consumer satellites. That means to keep up its growth rate, Starlink needs to find lots more customers because incremental customers are less lucrative than earlier ones. This also shows that it lacks real monopoly pricing power – while Starlink is faster than broadband, it’s not that much faster, and in the consumer market isn’t able to command much of a pricing premium.

Then there’s the addressable market. SpaceX bulls note that every broadband user on earth is a potential Starlink customer. But even if you take the ultimate steel man case and take the ten largest telcos on earth (AT&T, Verizon, Deutsche Telekom, Comcast, China Mobile, NTT, Vodafone etc) their annual profits combined are around US$110b.

It’s inconceivable that Starlink could take even half of that market given the incumbents’ rusted-on customer bases, massive infrastructure and often, diversified businesses. But even if Starlink took every single customer from every major telco on earth, the business would be worth around US$1.4 trillion – a significant discount on SpaceX’s current value.

That leaves the SpaceX rocket business, which notionally makes some EBITDA (US$1.1b in 2024 and US$653m in 2025) but which spends US$3b on capex annually. There’s no doubt SpaceX is best in class, but despite Elon’s Mars fever dreams, remains trapped in a relatively contained market. This year more than 75% of SpaceX’s payloads are shipping Starlink satellites into orbit (this is up from 64% in 2023).

The 25% of SpaceX flights that weren’t dedicated Starlink missions were shared among commercial telecom satellites, U.S. space force spy satellites and NASA astronaut/cargo runs to the ISS.

This means SpaceX can’t really be considered a separate operating business; it’s really part of Starlink, so it’s hard to give it much value.

So we have two very cool, but mostly valueless, businesses (in xAI and SpaceX) and one fantastic business (Starlink). The problem is Starlink is only a US$4b profit business (and that’s being generous) growing at 86% annually with shrinking ARPU. Even allowing for a significant Elon premium, this is maybe a US$300b business – if you just saw the financials and nothing else, you’d be closer to US$150b (or even less). That’s around 90% below the current market price. (None of this is investment advice and I don’t have a position in SpaceX stock.)

So how did this SpaceX bubble become so frothy? Partly there’s the Elon magic – Tesla has been similarly overvalued for almost a decade. But there were other reasons.

Most notably, the float raised only US$75b – this is tiny when compared to SpaceX’s total market value. So a small number of shares is acting as a false proxy. Then there was the controversial decision by the NASDAQ to alter its index rules to fast-track SpaceX into its flagship Nasdaq-100 index. This meant that SpaceX will join the index in just 15 days (rather than 3 months), forcing index funds and ETFs to buy the stock.

When you mix the Elon magic with a fast growing and much loved product and sprinkle on some financial alchemy, you get a business that is not only one of the most valuable ever, but could also drop 90%.

If you’re thinking this all feels like 1999 all over again, you’d be right.